Paul Vogt, Sr. Director, Product Management
Xyratex International, Ltd.
The Need for Serial Attached SCSI (SAS)
In order to understand the justification for the industry’s investment in a new I/O technology standard, it is useful to look at a typical parallel Ultra320 SCSI bus configurations. This type of configuration was a common Direct Attached Storage (DAS) implementation in an enterprise server configuration in the early 2000’s. Illustrated are two boot drives in the system as well as an additional Ultra320 SCSI JBOD (Just a Bunch of Disks) connected to the SCSI adapter in order to increase the capacity of the system.
Performance Roadmap Challenges with Ultra320 SCSI
Increasing the performance of an Ultra320 SCSI storage subsystem in order to keep up with the I/O requirements of a CPU complex with performance advancing at Moore’s Law rates uncovers the limitations of the parallel SCSI topology.
- The first issue to emerge is signal skew between the numerous data lines on the bus. The data bits don’t arrive at the other end of the cable at the same time because the signals travel at slightly different speeds as they travel down the cable. Because of the speed of the Ultra320 SCSI bus (6ns between the individual words) and the length of the cables (up to 12 meters long), up to six words of information can be in transit going down the cable before the first word ever reaches the destination. These skews can also change over time with heat and other harsh environmental potentials present in an enterprise server environment.
- The second issue to contend with when designing a performance increase in parallel SCSI is crosstalk. Crosstalk occurs in parallel interfaces because the signals are adjacent to each other in the cable. Each individual signal will interact with other signals through capacitive coupling, which leads to distortion between the signal pairs. As the signal edge rates increase, the potential for crosstalk also increases.
- The third issue encountered is equalization, or the need for equalization because these signals are driven in an NRZ (non-return to zero) format. Frequencies from direct current to the highest frequency components are contained within the signal transmission. Without proper equalization, the signals will contain frequency-related distortions as they travel down the bus.
- The final issue is one of reflections. As the signal travels down the bus and hits a discontinuity in the impedance such as a drive or a connector or even a printed circuit board (PCB) trace, a signal reflection will be created, traveling back down the cable plant. With cable lengths of up to twelve meters and up to six words of information in transit, these reflections cause further distortion of the signals traveling down the bus.
These four issues add up to a tremendous barrier to increasing the performance of the Ultra320 SCSI parallel standard.
Topology Scalability Challenges of Ultra320 SCSI
Another area challenging Ultra320 SCSI storage systems has been one of expandability. The SCSI standard was designed for up to 16 total devices on the bus. Expanding beyond 16 devices is difficult and expensive. Two options are discussed below, each with drawbacks:
- Parallel busses – one option is to add an additional Host Bus Adapter (HBA) with each JBOD storage enclosure. This parallels the bus architecture, but adds the incremental expense of an HBA.
- Cascaded JBODs – another option is to cascade an additional JBOD, but parallel SCSI (without an expensive bridge controller) can only have a total of 16 devices on the bus, one of which has to be the controller; limiting the total storage device count on the bus to 15.
The SAS Value Proposition
Serial Attached SCSI (SAS) includes significant improvements for both Direct Attached as well as Network Attached storage subsystems:
- Performance: While first generation SAS devices have a transfer rate of 3Gb/s which is similar to the performance that is possible from an Ultra320 SCSI interface, SAS has two significant improvements: One significant difference is the “wide-port” capability which will be covered in detail later. The other is the capability for full-duplex transfers which were not possible in parallel SCSI. Full-duplex links enable SAS devices to simultaneously transmit and receive, which effectively doubles the transfer rate across the link.
- Addressability: In terms of scalability SAS can address up to 16,000 devices which is greater than 1,000 times more than parallel SCSI.
- Topology: At first glance it appears that SAS topology is limited in proximity compared to parallel SCSI. However, SAS effectively supports longer distances because the cascading feature allows six-meter links between devices.
- Form Factor improvements: The small size of the serial connector is a fundamental enabler of the 2.5-inch disk drive form factor. SAS is leading the server transition to 2.5-inch form factor disk drives.
- Implementation costs: SAS has significant cost reduction over parallel SCSI due to reduced cable and connectors. Smaller cables and connectors also assist in reducing the cooling requirements within servers and cabinets by improving airflow within the system. Additional implementation cost savings are possible due to the nature of the serial protocol and shrinking silicon geometries. SAS brings significant cost advantages in terms of component costs by requiring fewer I/O pins to perform the equivalent function. Fewer I/O pins facilitates smaller packaging of ASICs and also benefits printed circuit board assembly costs by requiring fewer printed circuit board traces.
- Reliability, Availability, Serviceability (RAS) – fault isolation: SAS has better fault isolation than parallel SCSI because of the point-to-point architecture.
- RAS – fault tolerance: The dual-port nature of SAS enables more fault-tolerant topologies augmenting the point-to-point reliability improvements. Multiple controllers and cables can be connected to the storage device, allowing for controller/cable failover for enterprise storage systems.
- Flexible Device Attach: The SAS standard includes the ability to attach Serial ATA devices – enabling a single infrastructure to support low cost, high capacity hard drives – as well as performance-optimized enterprise drives. This flexibility reduces development and operation costs throughout the supply chain and provides customers with improved flexibility.
It is these benefits that justify the development of a new standard for storage infrastructures. Serial Attached SCSI builds on the proven SCSI command set and software infrastructure to provide a roadmap for the future which meets or exceeds the performance, scalability, cost and reliability needs of enterprise storage solutions.